another day, another problem. I’m trying to feed ptx files with scan into Reality Capture. I have 10 scans with position and rotation draft. Below is the screenshot from FaroScene. The scans are not aligned perfectly, but you can see the shape of the front of surikata statue.
The scans itself are quite detailed. Here is the image taken from RGB values of ptx files.
And finally if I import ptx files into Reality Capture there are several problems… Scans are converted to lsp cubemaps, but they are really small (140x140). After I try to reconstruct the model, resulting point cloud is completelly off (It doesn’t even resemble the original shape). My question is following: how is the resolution of the lsp files determined? Can the small resolution result in completelly different shape?
Is it posibble the RC is misinterpreting transformation of ptx (although the positions looks right)
My ptx file has 751 rows and 1026 columns. The fov it captures is significantly lower then 90° on lsp cubemap side. The ptx is sparse (which is not violation againts format specification). The problem is that tle lsp cubemap resolution is so tiny (140x140) that for the area which captures my ptx file a got like 12*12 pixels…thats ridiculous.
I understand you have good reasons to implement laser scans in cubemaps, but please let us have some control. Would it be posible to add some way to increase lsp cubemap resolution?
.lsp in 2D view should show the real resolution of the scan. If it is not a lot of data, would it be possible to share with us the .ptx file so that we can examine it? If yes, can you please send us download link via contact form? Thanks.
It seems to me that all the points from laser scans are reprojected to your own represantation (lsp cubemap) and the problem here is that algorithm for resolution determination is right now written for 360° scans. But my scans has fov of ~30°
After communication with RC support we came to solution:
My points were stored in “camera space” or “laser scan space” with forward pointing to positive z.
If axes were switched to xzy the resolution was determined correctly. Transformation matrix must be also modified so the world space coordinates of projected points stay the same.
There could be proper handling for this type of laser scans in the future.